CN104361102B - A kind of expert recommendation method and system based on group matches - Google Patents

Abstract

The invention discloses an expert recommendation method and system based on group matching, which belongs to the technical field of the Internet. The method includes: S1: obtaining the web page information of each expert in the expert list through a web crawler; S2: performing the web page information Extract to obtain the expert academic information of each expert; S3: Calculate the matching degree between each expert and the item to be matched according to the expert academic information; S4: Determine the matching degree and group matching model according to the dynamic programming algorithm as Experts recommended by the item to be matched. The present invention realizes expert recommendation through group matching, improves the efficiency of expert recommendation, and greatly reduces time overhead. In addition, when calculating the matching degree between each expert and the item to be matched, the relationship between each expert and the item to be matched is also considered. The degree of social relationship matching, thus effectively avoiding or preventing academic corruption when implementing expert recommendation.

Description

An expert recommendation method and system based on group matching

technical field

The invention relates to the technical field of the Internet, in particular to an expert recommendation method and system based on group matching.

Background technique

The evaluation efficiency and quality of scientific research projects have an important impact on the scientific research development level of a unit or even a country. As a fast and advanced evaluation method, online evaluation runs through the entire life cycle of a scientific research or engineering project from project establishment, application, organization, demonstration, evaluation, acceptance, reward to filing and other stages. Its purpose is to use computer and network systems to Replace traditional manual operations, thereby reducing review costs, improving work efficiency and review quality, and using electronic information systems to standardize the review process.

In recent years, the rapid development of new information technologies such as cloud computing, big data, recommendation systems, deep learning, and social networks has made intelligent network review possible, and the intelligent expert recommendation system is the core and difficulty of the entire network review process. The meaning of intelligence here is: the system can not only process and refine internal information (based on code, precise, structured), but also continuously gather external information (based on semantics, fuzzy, unstructured), through data accumulation. Experts themselves classify and evaluate to generate a more instructive intelligent expert database, thereby constructing more reasonable recommendation models and algorithms. However, the existing expert recommendation system has the problem of low expert recommendation efficiency, resulting in excessive time consumption.

Contents of the invention

In view of the above problems, the present invention provides an expert recommendation method based on group matching, the method comprising:

S1: Obtain the webpage information of each expert in the expert list through a web crawler;

S2: Extracting the webpage information to obtain expert academic information of each expert;

S3: Calculate the matching degree between each expert and the item to be matched according to the academic information of the expert;

S4: According to the matching degree and the group matching model, the expert recommended for the item to be matched is determined by a dynamic programming algorithm, and the group matching model is when the sum of the matching degrees of all recommended experts for the item to be matched reaches the maximum, The corresponding relationship between the item to be matched and the recommended expert.

Wherein, in step S1, the webpage information of each expert in the expert list is acquired through a web crawler according to the names of the experts in the expert list.

Wherein, step S2 specifically includes:

S201: Search the webpage information for the webpage information matching the expert name and work unit of the current expert, if not found, execute step S202, otherwise extract the expert's academic information from the first webpage information searched , and execute step S203, the list of experts includes: the name and work unit of each expert;

S202: Search for webpage information matching the expert name of the current expert from the webpage information, and extract expert academic information from the first searched webpage information;

S203: Use the experts whose academic information has not been extracted from the expert list as the current experts, and return to step S201.

Wherein, the academic information of the expert includes: the name of the expert, the work unit, keywords in the research field, the title of the paper and the author of the paper.

Wherein, in step S3, according to the academic information of the experts, the matching degree between each expert and the item to be matched is calculated by the following formula,

M _i,j ＝α*MK _i,j +β*MJ _i,j +γ*ML _i,j -δ*MS _i,j

Among them, M _i,j is the matching degree between expert i and project j to be matched, α, β, γ, and δ are all constants, and MK _i,j is the scientific research field keywords between expert i and project j to be matched Matching degree, MJ _i,j is the matching degree of journal conference labels between expert i and project j to be matched, ML _i,j is the academic level matching degree between expert i and project j to be matched, MS _i,j is expert The matching degree of social relationship between i and item j to be matched.

Wherein, the MS _i,j is calculated by the following formula,

in, is the weight value; δ _{i, j} is the correlation degree of the work unit between expert i and the vth expert, when the work unit is the same, the value of δ _{i, j} is 1, otherwise the value is 0; sp is expert i Papers that have cooperated with the vth expert; n is the number of paper authors; t _i is the weight of expert i; t _v is the weight of the vth expert; k is the applicant corresponding to the project j to be matched ; a is the number of project applicants corresponding to project j to be matched.

Wherein, the group matching model is:

in, c _i,j is the group matching matrix. When expert i recommends the jth item, the value of the i-th row and column j in the matrix is 1, otherwise the value is 0; m is the total number of items to be matched ; n is the total number of experts; ε is the maximum number of experts corresponding to each item to be matched; σ is the maximum number of items corresponding to each expert.

Wherein, in step S4, according to the matching degree and the group matching model, the expert recommended for the item to be matched is determined through a dynamic programming algorithm, specifically including:

S401: Determine the experts corresponding to each item to be matched according to the keywords in the research field, and sort the experts corresponding to each item to be matched according to the matching degree;

S402: Assign the experts with the highest matching degree to the corresponding items to be matched in turn until the number of experts corresponding to the items to be matched reaches the maximum value ε or the number of items corresponding to the experts reaches the maximum value σ.

The invention also discloses an expert recommendation system based on group matching, the system includes:

The web page acquisition module is used to obtain the web page information of each expert in the expert list through a web crawler;

An information extraction module, configured to extract the webpage information to obtain expert academic information of each expert;

A matching degree calculation module, configured to calculate the matching degree between each expert and the item to be matched according to the expert's academic information;

The expert recommendation module is used to determine the expert recommended for the item to be matched through a dynamic programming algorithm according to the matching degree and the group matching model, and the group matching model is the sum of the matching degrees of experts recommended for all items to be matched When the maximum value is reached, the corresponding relationship between the item to be matched and the recommended expert.

The present invention realizes expert recommendation through group matching, improves the efficiency of expert recommendation, and greatly reduces time overhead. In addition, when calculating the matching degree between each expert and the item to be matched, the relationship between each expert and the item to be matched is also considered. The degree of social relationship matching, thus effectively avoiding or preventing academic corruption when implementing expert recommendation.

Description of drawings

Fig. 1 is a flowchart of an expert recommendation method based on group matching in an embodiment of the present invention;

Fig. 2 is a structural block diagram of an expert recommendation system based on group matching according to an embodiment of the present invention.

Detailed ways

The specific implementation manners of the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

Fig. 1 is a flowchart of an expert recommendation method based on group matching in an embodiment of the present invention; referring to Fig. 1, the method includes:

S1: Obtain the webpage information of each expert in the expert list through a web crawler;

It should be noted that a web crawler is required to download a large amount of data in a general search engine, but the web crawler in step S1 only needs to visit a single site (that is, an academic website), and the rules are obvious, so there is no need to use distributed and webpage ranking and other technologies.

However, the target website needs to authenticate the identity, so we need to use the browser to parse the authentication process, and then simulate the browser to log in to the server for crawling. Generally speaking, the authentication mechanism of the website can be obtained by simulating cookies. After analyzing the target website, it is found that the target website is also authenticated through cookies. Therefore, first, manually log in to the website server to obtain a cookie with permission, and then copy it to the web crawler, and the web crawler uses this permission to crawl the webpage.

After analysis, when obtaining the webpage information of the expert, only need to change the parameter of the expert number range in the submitted http to obtain the corresponding webpage information. The obtained webpage information is stored in the hard disk in the form of html webpage, and then the webpage information is analyzed by using regular expressions to obtain expert information and build a basic expert database.

It can be understood that, in this step, the webpage information of each expert in the expert list is acquired through a web crawler according to the names of the experts in the expert list.

S2: Extracting the webpage information to obtain expert academic information of each expert;

It should be noted that, with the basic expert database, experts can be screened by name and work unit to extract expert academic information. In the process of extracting the academic information of experts, it is found that many experts in the expert list have the same name, so they can be distinguished by the work unit.

This step can be realized by the following steps S201-203:

S201: Search the webpage information for the webpage information matching the expert name and work unit of the current expert, if not found, execute step S202, otherwise extract the expert's academic information from the first webpage information searched , and execute step S203, the list of experts includes: the name and work unit of each expert;

S202: Search for webpage information matching the expert name of the current expert from the webpage information, and extract expert academic information from the first searched webpage information;

S203: Use the experts whose academic information has not been extracted from the expert list as the current experts, and return to step S201.

Optionally, the expert's academic information includes: expert's name, work unit, research field keywords, paper title and paper author.

S3: Calculate the matching degree between each expert and the item to be matched according to the academic information of the expert;

S4: According to the matching degree and the group matching model, the expert recommended for the item to be matched is determined by a dynamic programming algorithm, and the group matching model is when the sum of the matching degrees of all recommended experts for the item to be matched reaches the maximum, The corresponding relationship between the item to be matched and the recommended expert.

When implementing expert recommendation, experts in related fields are needed to review projects, so the matching degree between experts and projects is positively related to the domain correlation between experts and projects; in order to avoid and prevent academic corruption, if the project applicant and review experts There is a social relationship between experts, such as co-authored papers, in the same work unit, etc., then the matching degree between experts and projects is inversely proportional to the social relationship between experts and projects. Therefore, the steps In S3, according to the academic information of the experts, the matching degree between each expert and the item to be matched is calculated by the following formula,

M _i,j ＝α*MK _i,j +β*MJ _i,j +γ*ML _i,j -δ*MS _i,j

Among them, M _i,j is the matching degree between expert i and project j to be matched, α, β, γ, and δ are all constants, and MK _i,j is the scientific research field keywords between expert i and project j to be matched Matching degree, MJ _i,j is the matching degree of journal conference labels between expert i and project j to be matched, ML _i,j is the academic level matching degree between expert i and project j to be matched, MS _i,j is expert The matching degree of social relationship between i and item j to be matched.

In order to calculate the matching degree of the social relationship, optionally, the MS _i,j is calculated by the following formula,

in, is the weight value; δ _{i, j} is the correlation degree of the work unit between expert i and the vth expert, when the work unit is the same, the value of δ _{i, j} is 1, otherwise the value is 0; sp is expert i Papers that have cooperated with the vth expert; n is the number of authors of the paper; t _i is the weight of expert i (the weight can be determined according to the order of the authors of the paper); t _v is the weight of the vth expert (the The weight can be determined according to the order of the authors of the paper); k is the serial number of the project applicant corresponding to the project j to be matched; a is the number of project applicants corresponding to the project j to be matched.

Optionally, assume that the keywords of expert i’s scientific research field are expressed as vectors <K _i,1 ,K _i,2 ,K _i,3 ,...,K _i,N >, and its weight (actually keyword frequency) It is expressed as a vector <w _i,1 ,wi _,2 ,wi _,3 ,...,wi _,N >; the keywords of the scientific research field of project j are expressed as a vector <K _j,1 ,K _j,2 , K _j,3 ,...,K _j,N >, its weight (actually keyword frequency) is expressed as a vector <w _j,1 ,w _j,2 ,w _j,3 ,...,w _{j, N} >. Using the content-based vector recommendation algorithm, define the keyword matching degree MK _i,j in the scientific research field as:

Here R(K _i,x ,K _j,y ) represents the similarity between keywords K _i,x and K _j,y of two scientific research fields (R=Resemblance).

Since keywords in the field of scientific research are usually not particularly accurate when extracting them, when calculating R(K _i,x ,K _j,y ), first calculate the edit distance between two scientific research keywords: Levenshein distance, edit distance It refers to the minimum number of editing operations required to convert from one string to another between two strings. Permissible editing operations include replacing one character with another, inserting a character, and deleting a character. Assuming that the edit distance is d, and the length of the longest word among the two scientific research keywords is max, then the similarity is 1-d/max.

Assume that expert i’s journal conference labels are expressed as vectors <J _i,1 ,J _i,2 ,J _i,3 ,...,J _i,N >, and their weights (actually label frequencies) are expressed as vectors <w _{i ,1} ,wi _,2 ,wi _,3 ,...,wi _,N >; the journal conference label of project j is expressed as a vector <J _j,1 ,J _j,2 ,J _j,3 ,.. .,J _j,N >, its weight (actually label frequency) is expressed as a vector <w _j,1 ,w _j,2 ,w _j,3 ,...,w _j,N >. Using the tag-based vector recommendation algorithm, define the journal conference tag matching degree MJ _i,j as:

Here I(J _i,x ,J _j,y ) indicates whether the conference labels J _i,x and J _j,y of two journals are the same (I=Identity).

It should be noted that since journal conference labels are generally very accurate, unlike R(K _i,x ,K _j,y ) between 0 and 1.0, I(J _i,x ,J _j,y ) when Journal conference labels are equal to 1 and 0 if not equal.

Assume that the academic level vector of expert i is <unit level, professional title, scientific research project scale>, where the unit level is the level of the university: such as 985, 211, general undergraduate, junior college, etc.; professional title: such as master tutor, doctoral tutor, Cheung Kong Scholars, academicians, etc.; scale of scientific research projects: applications have been completed, including completed and ongoing national scientific research projects: such as 863 projects, etc., and the evaluation index is the amount of scientific research funds;

For a set represented by such a set of vectors, the k-means clustering method is used to cluster the student level of experts. First, k representative experts are selected, and experts similar to the representative experts are put into a class, so that Put experts into k categories. When calculating the academic level similarity of two experts, if the two experts belong to the same category, then the similarity is 1, and if the two experts are not in the same category, then the similarity is 0.

Then we can use such a set of vectors V ₁ , V ₂ , V ₃ , ..., V _p to represent the academic level of applicants for each project (assuming that there are p applicants in project j). When calculating the academic matching degree between an expert and a project, the similarity between the expert and each project applicant can be calculated separately, then summed, divided by the number of applicants, and the degree of similarity between the project and the expert can be obtained similarity.

Optionally, the group matching model is:

in, c _i,j is the group matching matrix. When expert i recommends the jth item, the value of the i-th row and column j in the matrix is 1, otherwise the value is 0; m is the total number of items to be matched ; n is the total number of experts; ε is the maximum number of experts corresponding to each item to be matched; σ is the maximum number of items corresponding to each expert.

In order to facilitate the determination of experts recommended for the item to be matched, optionally, in step S4, the expert recommended for the item to be matched is determined through a dynamic programming algorithm according to the matching degree and the group matching model, specifically including :

S401: Determine the experts corresponding to each item to be matched according to the keywords in the research field, and sort the experts corresponding to each item to be matched according to the matching degree;

S402: Assign the experts with the highest matching degree to the corresponding items to be matched in turn until the number of experts corresponding to the items to be matched reaches the maximum value ε or the number of items corresponding to the experts reaches the maximum value σ.

The present invention also discloses an expert recommendation system based on group matching. Referring to FIG. 2, the system includes:

The web page acquisition module is used to obtain the web page information of each expert in the expert list through a web crawler;

An information extraction module, configured to extract the webpage information to obtain expert academic information of each expert;

A matching degree calculation module, configured to calculate the matching degree between each expert and the item to be matched according to the expert's academic information;

The expert recommendation module is used to determine the expert recommended for the item to be matched through a dynamic programming algorithm according to the matching degree and the group matching model, and the group matching model is the sum of the matching degrees of experts recommended for all items to be matched When the maximum value is reached, the corresponding relationship between the item to be matched and the recommended expert.

The system also includes modules, sub-modules, units, and sub-units for realizing each step of the above-mentioned method, and in order to avoid repeated description, details are not repeated here.

The above embodiments are only used to illustrate the present invention, but not to limit the present invention. Those of ordinary skill in the relevant technical field can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, all Equivalent technical solutions also belong to the category of the present invention, and the scope of patent protection of the present invention should be defined by the claims.

Claims (5)

1. An expert recommendation method based on group matching, the method comprising:

s1: acquiring webpage information of each expert in the expert list through a web crawler;

s2: extracting the webpage information to obtain expert academic information of each expert;

s3: calculating the matching degree between each expert and the item to be matched according to the expert academic information;

s4: determining experts recommended for the items to be matched through a dynamic programming algorithm according to the matching degree and a group matching model, wherein the group matching model is a corresponding relation between the items to be matched and the recommended experts when the sum of the matching degrees of all the recommended experts of the items to be matched reaches the maximum;

the expert academic information comprises: expert names, work units, research field keywords, paper names and paper authors;

in step S4, determining, according to the matching degree and the group matching model, an expert recommended for the item to be matched through a dynamic programming algorithm, specifically including:

s401: determining the experts corresponding to each item to be matched according to the research field keywords, and sequencing the experts corresponding to each item to be matched according to the matching degree;

s402: sequentially distributing the experts with the highest matching degree to the corresponding items to be matched until the number of the experts corresponding to the items to be matched reaches a maximum value epsilon or the number of the items corresponding to the experts reaches a maximum value sigma;

in step S3, the matching degree between each expert and the item to be matched is calculated according to the expert academic information through the following formula,

M _i,j ＝α*MK _i,j +β*MJ _i,j +γ*ML _i,j -δ*MS _i,j

wherein M is _i,j For the matching degree between the expert i and the item j to be matched, alpha, beta, gamma and delta are constants, MK _i,j For expertsDegree of matching, MJ, of scientific research field keywords between i and item j to be matched _i,j For the matching degree of the labels of the periodical meeting between the expert i and the item j to be matched, ML _i,j For the academic level matching degree between the expert i and the item j to be matched, MS _i,j Matching degree of social relationship between the expert i and the item j to be matched;

the MS _i,j The calculation is carried out by the following formula,

wherein, is a weighted value; delta. For the preparation of a coating _i,j Is the correlation degree of the working units between the expert i and the expert v, when the working units are the same, delta _i,j The value of (1) is 1, otherwise the value of (0) is obtained; sp is a thesis that the expert i cooperates with the nth expert; n is the number of paper authors; t is t _i The weight occupied by expert i; t is t _v The weight occupied by the vth expert; k is the serial number of the project applicant corresponding to the project j to be matched; a is the number of the item applicants corresponding to the item j to be matched.

2. The method as claimed in claim 1, wherein in step S1, web page information of each expert in the expert list is obtained through a web crawler according to the names of the experts in the expert list.

3. The method according to claim 2, wherein step S2 specifically comprises:

s201: searching the web page information matched with the expert name and the working unit of the current expert from the web page information, if the web page information is not searched, executing the step S202, otherwise, extracting expert academic information from the searched first web page information, and executing the step S203, wherein the expert list comprises: the expert names and work units of each expert;

s202: searching webpage information matched with the expert name of the current expert from the webpage information, and extracting expert academic information from the searched first webpage information;

s203: the experts from which the expert academic information is not extracted in the expert list are taken as the current experts, and the process returns to step S201.

4. The method of claim 1, wherein the group matching model is:

wherein,c _i,j matching a matrix for the group, wherein when an expert i recommends a jth project, the jth column of a corresponding ith row in the matrix takes a value of 1, and otherwise, the jth column takes a value of 0; m is the total number of items to be matched; n is the total number of experts; epsilon is the maximum value of the number of experts corresponding to each item to be matched; sigma is the maximum value of the number of items corresponding to each expert.

5. An expert recommendation system based on group matching, the system comprising:

the webpage acquisition module is used for acquiring webpage information of each expert in the expert list through a web crawler;

the information extraction module is used for extracting the webpage information to obtain expert academic information of each expert;

the matching degree calculation module is used for calculating the matching degree between each expert and the item to be matched according to the expert academic information;

the expert recommending module is used for determining the experts recommended for the items to be matched through a dynamic programming algorithm according to the matching degree and a group matching model, and the group matching model is the corresponding relation between the items to be matched and the recommended experts when the sum of the matching degrees of all the recommended experts of the items to be matched reaches the maximum;

the expert academic information includes: expert names, work units, research field keywords, thesis names and thesis authors;

the expert recommendation module is specifically configured to:

determining the experts corresponding to each item to be matched according to the research field keywords, and sequencing the experts corresponding to each item to be matched according to the matching degree;

sequentially distributing the experts with the highest matching degree to corresponding items to be matched until the number of the experts corresponding to the items to be matched reaches a maximum value epsilon or the number of the items corresponding to the experts reaches a maximum value sigma;

calculating the matching degree between each expert and the item to be matched according to the expert academic information through the following formula,

M _i,j ＝α*MK _i,j +β*MJ _i,j +γ*ML _i,j -δ*MS _i,j

wherein M is _i,j For the matching degree between the expert i and the item j to be matched, alpha, beta, gamma and delta are constants, MK _i,j Matching degree of keywords, MJ, in the scientific research field between expert i and item j to be matched _i,j For the matching degree of the labels of the periodical meeting between the expert i and the item j to be matched, ML _i,j For the academic level matching degree between the expert i and the item j to be matched, MS _i,j Matching degree of social relationship between the expert i and the item j to be matched;

the MS _i,j The calculation is carried out by the following formula,

wherein, is a weighted value; delta. For the preparation of a coating _i,j Is the correlation degree of the working units between the expert i and the expert v, when the working units are the same, delta _i,j The value of (1) is 1, otherwise the value of (0) is obtained; sp is a thesis that the expert i cooperates with the nth expert; n is the number of paper authors; t is t _i The weight occupied by expert i; t is t _v The weight occupied by the vth expert; k is the serial number of the project applicant corresponding to the project j to be matched; a is the number of the item applicants corresponding to the item j to be matched.